Avoiding moderate BIASES and moderate RANDOM ERRORS

1
Avoiding moderate BIASES andmoderate RANDOM
ERRORS

• Avoid MODERATE systematic errors (biases)
• Proper RANDOMISATION and analysis by ALLOCATED
  treatment
• Chief emphasis on OVERALL results (rather than
  on particular subgroups)
• META-ANALYSES of all relevant studies(rather
  than emphasising particular studies)
• Avoid MODERATE random errors
• LARGE numbers of outcomes in new studies (with
  streamlined methods to facilitate recruitment)
• META-ANALYSES of all relevant studies (yielding
  largest number of major outcomes)

2
Different SIZE of effects in observational
versus randomised studiesBeta-blocker use and
death following MI
3
Different DIRECTION of effects in observational
versus randomised studiesAntihypertensive
therapy and coronary events
4
Total mortality as an endpoint

• Few drugs influence multiple causes of deaths, so
  effects on total mortality modest unless one
  cause of death dominates (eg, heart failure)
• Unless this is so, detecting plausible effects on
  total mortality requires ENORMOUS trials
• In any case, effects on total mortality not
  generalisable to different populations
  cause-specific mortality is more informative